基于相似度辅助决策的带宽自适应跟踪算法  

Tracking algorithm with adaptive bandwidth based on assistant decision-making of similarity metric

作　　者：刘翔[1] 侯志强[1] 余旺盛[1] 黄安奇[1] 

机构地区：[1]空军工程大学信息与导航学院,西安710077

出　　处：《计算机应用研究》2015年第9期2849-2852,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61175029);陕西省自然科学基金资助项目(2011JM8015)

摘　　要：针对传统窗宽固定不变的Mean-Shift跟踪算法不能实时地适应目标尺寸大小变化这一问题,提出了一种基于目标相似度辅助决策的带宽自适应跟踪算法。首先利用目标与背景的特征显著性,提高跟踪算法空间定位准确性;然后利用局部穷搜索的方法,计算目标模型与每一帧目标跟踪中心点附近一定区域的相似性;最后通过统计分析前后帧相似像素点数目变化,确定目标尺度变化情况,从而建立一种自适应更新带宽准则,提高算法对目标尺度变化的自适应性。实验结果表明,改进的算法可以有效地提高Mean-Shift跟踪算法空间和尺度定位准确性。This paper researched the problem that the traditional Mean-Shift algorithm could not track a size-changing object effectively, and proposed a new algorithm based on assistant decision-making of object similarity metric, to estimate the scale and orientation of a tracking window with adaptive bandwidth. Firstly, it adopted the saliency of object and background to im- prove tracking accuracy for orientation, and then employed the local exhaustive search to compute the similarity metric between object model and the certain region where was around the tracking center in each frame. Finally, it determined the object scale variation by similar pixel amounts. What＇ s more, it defined a novel bandwidth criterion for improving adaptability in tracking bandwidth. The experimental results prove that the present method can improve the tracking accuracy effectively in orientation between space and scale.

关 键 词：视觉跟踪 均值漂移算法 显著性加权 自适应带宽 

分 类 号：TN919.8[电子电信—通信与信息系统]